WO2015101418A2

WO2015101418A2 - Methods and devices for doppler shift compensation in a mobile communication system

Info

Publication number: WO2015101418A2
Application number: PCT/EP2014/003486
Authority: WO
Inventors: Thorsten Herfet; Nasimi ELDAROV
Original assignee: Universität des Saarlandes
Priority date: 2013-12-30
Filing date: 2014-12-30
Publication date: 2015-07-09
Also published as: EP3090519A1; WO2015101672A1; EP3090519B1

Abstract

A method for compensating Doppler shifts (v) in a mobile communication system comprises the steps of: receiving a mobile communication signal (y); transforming the signal in order to compensate a Doppler shift (v_p) of at least one propagation path (p) of the signal (y); and outputting the compensated signal. A matching pursuit (MP) algorithm is used for estimating Doppler shifts of propagation paths (p) of the signal.

Description

Methods and Devices for Doppler Shift Compensation

in a Mobile Communication System

INTRODUCTION In the field of wireless communications, OFDM has received much recent attention because of its high spectral efficiency, simple equalization, low complexity FFT based implementation, among other aspects. Therefore, many systems e.g., Digital Video Broadcasting (DVB), Digital Audio Broadcasting (DAB), Long Term Evolution (LTE), Worldwide Interoperability for Microwave Access (WiMAX) etc. are based on OFDM [l], [2], [3], [4]. However, there exist obstacles to high mobility in such systems, as despite of its nice features, OFDM is very sensitive to Doppler shift, where subcarriers are shifted causing an orthogonality loss in the receiver. This phenomenon can also be found in literature as an intercarrier interference (ICI). As an example, in LTE, in the Doppler shift scenario, the high performance can be supported at speeds between 15 to 120 km/h depending on the frequency band and the direction of the movement. Although OFDM can cope with inter symbol interference (ISI) with the help of a cyclic prefix (CP) efficiently, in presence of the ICI this becomes inefficient.

PRIOR ART

Although there have been many methods to cope with ICI, all are similar in nature. In general, the proposed methods have two parts. The first part is the estimation of the channel state information and second is the inversion of the channel matrix and equalization of the symbols. Among these related works, A. Nakamura et.al. (A. Nakamura and M. Itami, Zero-forcing ICI canceller using iterative detection for mobile reception of OFDM,ICCE 2014, 2014,153-154) proposed a zero-forcing ICI canceller which reduces the complexity of channel matrix inversion 0(N³j by skipping some non-diagonal elements. However this approach still becomes very complex. Besides them, Y.H. Zhang et.al. proposed an optimization problem which can cancel ICI recursively (Y.H. Zhang, W.-S. Lu, T.A. Gulliver, A Successive Intercarrier Interference Reduction Algorithm for OFDM Systems,IEEE International Conference on Communications, ICC 2007., 2007, 2936-2941). Apart from deriving the channel matrix, the matrix can be inverted with a complexity of 0(,N^l) and parameters updated recursively which makes it quite hard for the systems with a high FFT size. Another ICI cancellation algorithm has been proposed as a partial minimum mean square error estimation (MMSE) which has a complexity varying between 0(N²) and 0(J¥³) depending on the parameters apart from the channel estimation [W. Hongmei, C. Xiang, Z. Shidong and Y. Yan, A low-complexity ICI cancellation scheme in frequency domain for OFDM in time-varying multipath channels, PIMRC 2005., 2005, 1234-1237 Vol. 2]. Both decrease the complexity by sacrificing the accuracy. Recently, T. Hrycak et.al. proposed a low-complexity ICI cancellation algorithm based on basis expansion model (BEM) coefficients (T. Hrycak, S. Das, G. Matz and H.G. Feichtinger, Low Complexity Equalization for Doubly Selective Channels Modeled by a Basis Expansion, IEEE Transactions on Signal Processing, 2010). However, the accuracy of BEM model is not satisfactory due to the modeling error, which decreases the final performance of the ICI cancellation.

Existing solutions are either very complex or inaccurate that make all of them inefficient in practice.

It is therefore an object of the invention to compensate Doppler shifts in mobile telecommunication signals efficiently and with greater accuracies. BRIEF SUMMARY OF THE INVENTION

This object is achieved by a method and a receiving device for compensating Doppler shifts according to the independent claims. Advantageous embodiments are defined in the dependent claims.

According to an aspect of the invention, a method for compensating Doppler shifts comprises the steps of receiving a signal (y), transforming the signal in order to compensate a Doppler shift (n_p) of at least one propagation path (p) of the signal (y) and outputting the compensated signal.

The step of transforming may comprise applying a Fast Fourier Transform (FFT) to the signal. The signal may be an OFDM signal.

The Doppler shift may be compensated based on an inverted channel matrix of the communication system. The channel matrix may include a representation of the Doppler shift (n_p) of the at least one propagation path (p) of the signal (y). The channel matrix may be estimated from the received signal. The channel matrix may be estimated using a matching pursuit (MP) algorithm. Channel coefficients may be determined using least squares (LS) minimization. The channel matrix may further include a representation of at least one of the attenuation, phase shift and time delay of the propagation path. A stopping criterion for the matching pursuit algorithm may take a distance between channel coefficients for different numbers of paths into account. A stopping criterion for the matching pursuit algorithm may further take a signal-to-noise-ratio (SNR) into account. BRIEF DESCRIPTION OF THE FIGURES

These and other aspects and advantages of the present invention will become more apparent when considering the following detailed description of embodiments of the invention, in connection with the drawing in which

Fig. l shows a flow chart of a method for compensating Doppler shifts in an

OFDM system according to an embodiment of the invention.

Fig. 2 shows a flow chart of a method for estimating a channel matrix according to an embodiment of the invention.

Fig. 3 is an illustration of a search for a maximum projection matching the target signal. Fig. 4 shows the performance of the Doppler compensation in comparison to the l-tap frequency domain equalizer, BEM based equalizer (which is one of the state-of-art methods) and that without Doppler shift compensation case. Fig. 5 shows the FEC threshold with red line after which BER becomes zero in case FEC is activated.

Fig. 6 introduces the adaptive stopping criterion which can reduce the computation after FEC threshold similar to the adaptive dictionary.

Fig. 7 shows a schematic MISO transmission and reception system. shows MISO scattering pilots according to an embodiment of the invention.

Fig. 9 shows a comparison of different channel estimation algorithms according to different embodiments of the invention.

Fig. 10 shows a comparison between the proposed and conventional MISO. shows a comparison between SISO and MISO.

DETAILED DESCRIPTION

Figure l shows a flow chart of a method for compensating Doppler shifts in an OFDM system according to an embodiment of the invention.

In step no, after reception, the received symbols are converted from serial to parallel and entered into the FFT block, where they are processed in step 120.

The mathematical description of the FFT for I = 0, ,„ N— 1 can be written as

N ~ 1

X(t) = J _*Cfr) exp(-/2^) ⁽¹⁾ In step 130, the inverse of a channel matrix H is calculated.

As the channel model, a linear time-variant (LTV) multipath propagation model is used, which can also be called a double selective channel where both time and Doppler shifts occur due to several reasons e.g., the mobility of either the transmitter or receiver, frequency offsets, oscillator drifts, multipath propagation, etc. Considering that OFDM is more sensitive to Doppler shifts than conventional frequency division systems, there is a need to compensate these effects for OFDM to achieve a high mobility.

In real wireless transmission scenarios, the transmitted signal will be faded, time delayed, phase and Doppler shifted. Specifically, Doppler shifts occur when at least one of the objects is moving in mobile environments. For the jj-th path, where v> is the speed of the object, φ_ρ is the angle of arrival, f_c is the carrier frequency and c is the speed of light.

According to the invention, the Doppler shift is considered as a constant within given symbol duration. In other words, the Doppler shift in the beginning of the symbol will be very close to the Doppler shift at the end of the symbol. The second assumption is the narrowband OFDM case where bandwidth is very small in comparison to the carrier frequency, B « f_c [5]. This assumption does not degrade the accuracy. According to the invention, these assumptions allow that Doppler shifts can be compensated efficiently with low complexity.

As a mathematical model of LTV, the channel transfer function can be denoted with h , t), the number of propagation paths with P, and input signal with x(t}. It is assumed that the guard interval T_CI is long enough for the maximum delay in the channel. The impulse response can be formulated as follows

where attenuation, time delay and Doppler shift are denoted with h_p, τ_ρ and {_p for p- th path. The received signal y(t) will be a circular convolution of the transmitted x(t) signal and impulse response ¾(τ, t)

(3)

> £} = h(T:.t)®x(t) + w(t) where <® denotes circular convolution and w(i) is a noise in the channel which is modeled as additive white Gaussian noise (AWGN) but will be omitted here for simplification.

As the received signal y(t) enters into the FFT block, the received signal is transformed into the frequency domain. Considering that the circular convolution becomes a multiplication, the received signal can be written as

where F(h(t)) denotes the Fourier transform of h(t). When equation (2) is written in (4) one obtains

If the Doppler component β*^Ζπν?* is denoted with d_p t) and rewritten inside (4), the received signal Y(f^~) will be simplified as follows

y(,f)

By the Fourier transform, δ(τ - τ_ρ) will be converted to e ^J2mf ^ , (t) ) to D_p (f) and the multiplication becomes a circular convolution as a result

⁼Σ h(^D?(fy®*~^S *) V> (7) p=C<

For further simplifications, A_e ^_j2rT/T? is denoted with fiL and rewritten as follows

here, for every path the circular convolution can be written as a multiplication of the circulant Doppler matrix and the diagonal delay matrix D (f)®H_p(f) = · H_p . And finally, by denoting Σζ ' fi wit if in a matrix form (8) can be rewritten as follows

Y = H^■ X ^

DOPPLER SHIFT COMPENSATION

In order to compensate Doppler shifts, one has to find if^-1 since the main task is to estimate X signal with a linear equalization as 2 = M^'1 · Y-

According to an embodiment of the invention, the channel transfer function for the time-variant channel can be inverted in the frequency domain based on a FFT, in just 0(NlogN^~) operations in case the assumptions that were made in the previous section hold.

According to equation (8), the channel response in the frequency domain written as

The matrix H_CP can be written as

The above matrix is a circulant matrix, and it was proven that this kind of matrices can be inverted based on the FFT with just 0(j¥ logiV) operations (P. Feinsilver, Circulants, inversion of circulants, and some related matrix algebras, Linear Algebra and Appl., 56, 29-43, 1984 and T. Hrycak, S. Das, G. Matz and H.G. Feichtinger, Low Complexity Equalization for Doubly Selective Channels Modeled by a Basis Expansion, IEEE Transactions on Signal Processing, 2010). Hence, complex circulant matrices can be diagonalized in terms of a Fourier matrix where N permutation matrix can be denoted with Π(= Π_¾-) and for the elements γ = [f/_p0* _"■ _> ^tne circulant matrix H_cp can be written as a polynomial relationship of _Κ(Π) as follows

Λ7- 1

(12) k—Q

If the Fourier transformation's primitive n-th root of unity is denoted as ω := e ^$ , the diagonal matrix Ω(= Ω_γ) can be formulated as

(13)

Π = diag it, ω, ω² , ω^{Λ 1} ) After constructing the Fourier matrix F{= F_N) and its Hermitian adjoint F* based on ω (= ω_Λί) the characteristic polynomial and H_cp can be shown in terms of Fourier representation as

¾_p( ) = *P (n) ⁽¹⁴⁾

The Eigenvalues of _cp(y) are P_r(m^2:), which makes M_cp invertible in 0{N logN): operations. In practice, there are several algorithms proposed for this process. Inspired by the sparse linear equations and sparse least squares (LSQR) based algorithm, one may also use the LSQR based algorithm to do the channel equalization.

CHANNEL RESPONSE CONSTRUCTION

According to an embodiment of the invention, a MP algorithm which is greedy in nature and accurate enough for the proposed Doppler compensation can be used to estimate the channel response, if it is not known a priori. One may either implement Basic MP (BMP) or least squares MP (LS-MP) which chooses the most dominant taps based on a dictionary (W. Li and J. C. Preisig, Estimation of Rapidly Time-Varying Sparse Channels, IEEE Journal of Oceanic Engineering, 2007, 32, 4, 927-939).

The channel dictionary D has elements of φ ,/ = Ί, .,.,Μ where M = K · L for K number of time shifts and I number of Doppler shifts in the dictionary. 5 denotes the number of the elements which have to be searched in the dictionary. At each iteration, the algorithm matches one of the elements of the dictionary based on the maximum rank-one projection and then subtract it from the residual signal b .

In total, the algorithm can iterate up to P possible paths until the residual goes to zero. Indexes and coefficients are denoted with l_p and c_p. Figure 2 shows a flowchart of a method for estimating a channel response according to an embodiment of the invention.

In step 210, the initial residual signal b₀ is set by calculating b_Q - 4> Y for / = 1, ... ,S. Also, the l_p vector is added to store selected projection indexes.

In step 220 an iteration is started for each path p = 1, ... ,P.

In step 230, a jj-th maximal projection i._p is calculated as

In step 240, i_v is stored in / - [l_v__ir £_.].

In step 250, coefficients are calculated as

In step 260, the residual vector is updated as

(17)

P>J

In step 270, a stopping criterion is checked. If the criterion is met, the channel coefficient estimates are output in step 280. Otherwise, the method braches back to begin a new iteration with another path in step 220.

In the algorithm above, equation (16) calculates the coefficients for the BMP algorithm.

However, for the LS-MP algorithm the coefficients are calculated based on LS as follows arg mm l — _s-_ac ^* = [D^ (18)

where D_i>p J. This is more accurate since the coefficients inside the MP algorithm are calculated using LS.

Phase shifts are calculated implicitly when coefficients are calculated. At the end of the algorithm, one gets c_p which can be considered as h_p for the p-th. path. Time delay r_p and Doppler shift v_p are derived from the selected dictionary element i_p at each iteration.

Running MP algorithms for all elements of the dictionary can be very complex. According to the invention, the complexity may be decreased by a novel heuristic search algorithm taking advantage of the two dimensions in the dictionary which are time delay and Doppler shift, while keeping the same quality.

One can first search for the delay in the dictionary by keeping the Doppler shift constant. After finding the delay, one can search for the Doppler element given the found delay.

Figure 3 shows an example of a dictionary search according to an embodiment of the invention. More particularly, the search algorithm may be described as a tree search where the root elements have 10 precision 6 elements starting from o up to 50 με· (O a , .,._.,50 us).

Assuming that the true delay is 1.245 us, the algorithm will search for 6 root delay elements and find the delay element τ_{ί μ2} = 10 μς as the root element in the tree. After that, as the next step, the algorithm searches for 9 delay elements below and 9 delay elements above the selected τ.±_0μ3 - 10 μ$ with a precision of 1 pis as 19 branches of the tree in total (1 μ , 2 LIS, ... , 9 us, 10 pis, 11 pis, 13 pis_r 19.pis). The algorithm will choose T_1jUS = 1 μ in this case. Up to here, the total elements that have to be searched are 24. The next step will be searching the branches of the tree around τ_ίμ3 = 1 5 to find T.^_IFT3 which will be = 1.2 ps. To find the final delay the algorithm continues in the same way two more times with ten times more precision by searching 18 elements at each step, and find the final 1.245 pes delay in just 36 operations. After that, for τ = 1.245 pts, one searches for the Doppler shift in the similar way in 20 + 10 operations. For Doppler shift, the branches are restricted to 10 elements instead of 18. In total, 5 will be 24 + 36 + 20 = 90. If one wants to just stop at precision of Q.ips delay 5 will reduce to 54. P is the maximum number of the iteration in MP algorithm having 32 as maximum. However, in a simulation done by the inventors it did not go over 40.

A dynamic dictionary whose elements are computed and then kept in memory may be used to reduce the computation. In this case the complexity will be ■ log jV_pitot) for each element.

If the threshold is denoted as T„ square of the norm of the estimated coefficients difference, the algorithm will stop when

-SNSfsea

^■. On the left side of the condition, T will be be bigger than 1 when new

2 entered element modifies all the coefficients that have been decided. In this case, since a new element will not bring any new information it can be stopped.

The maximum projection is R_p

As may be seen on the right side of the condition the stopping criterion is adaptive to SNR which makes the algorithm robust to noise.

COMPUTATIONAL COMPLEXITY

In case N is the number of carriers in an OFDM symbol equal to 8K, AL_iiet is the number of pilots in the OFDM symbol equal to 576, p is the index of the path and the maximum number of the iteration in the MP algorithm up to P. In general, the dictionary will have M = K · L elements where the first multiplier is the number of possible delays and can be calculated as K = = 56 με/ 1 ns where is a division of maximum delay (T_max = T_C1) to granularity of delays (Ins). The second multiplier is the number of the Doppler elements and can be calculated as

L = ^⁵³^ = 200 Hz/1 Hz where is a division of maximum Doppler shift (200 Hz) to the Doppler intervals (1H∑). The complexity of FFT operations is OiN logN). This part is very important since this was very complex before similar to 0(W³). The complexity of the MP algorithm can be reduced by the search heuristic described in relation to figure 3. Considering that the number of the search steps is S, the complexity of the MP can be written as 0(S^■ P²) where 5 can be 90 at most. The Doppler shift compensation according to the invention can be exemplified with a simple 2-path model with angles o° and 45⁰ and delays o and 1.6 β. The carrier frequency and speed are 862 MHz and 250 km/h which translates to a Doppler shift of 200 Hz. Here, angles mean that there is only one moving receiver seeing paths under those angles. For simplicity, assume an FFT of size N = 4, sampling interval 224 μβ and X = [-1,1, - 1/1]. First, the IFFT is applied to X and one gets x = - [0,0,— lM where * enters into the LTV channel. In the LTV channel paths are denoted with ii and h₂ vectors which represent the channel impulse response for every symbol n = 1, in the channel and can be written after adding the delays as

h₂ = [-0.0000 4- 0.00284 -0.0028 - 0.002-8 0.5 + 0.00284 0.0028 - 0.0028?]

After adding delays, Doppler shifts are added to the paths. Doppler shift vectors can be constructed using (7) and written as

D. = [0.9606 4- 0.27784 0.8457 -r 0,5337 0.6641 - 0.747640.4303 4- 0.90271] ^' D_z - (0.9803 ÷ 0.19774 0.9218 + 0.387740.8270 4- 0.562340.6995 4^» 0.71471]

Finally, Ji-i and ft 2 can be summed and h may be written for the channel as h = [-0.0006 + 0.00274 -0.0015 - 0.00374 0.7439 - 0.65724 0.0040

- 0.00014] Where the received signal y enters into the FFT block and one gets Y as

Y = [-0.7459 - 0.6S644 0.7482 ~ 0.64904 -0.7408 - 0.66364 0.7408

4- 0.65 9J]

Finally, the Y signal can be equalized by calculating the inverse of H which can be written as

One may calculate H by applying FFT to (14) and write

Hi®T0i) = [0.3321 +- 0.373840,3321 4- .373840.3321 + G.37384 G.3321

4- 0.3738*1

H_? HD_z) = [0.4138 4- 0.28.2640.4162 ÷ 0.275240.4088 + 0.2898 0.4087

4- 0.2861Ϊ1

The sum of them is

H = (0.7459 4· 0.6564 0.7482 4- 0.64904 0.7408 4 0.663640.7408 4- 0.6S99f] The compensated signal is X - Y./H = [—1,1, - 1,1] which is equal to X.

A prototype of the invention was implemented in DVB-T2, a common simulations package (DVB-T2 CSP) as a 32 bit Matlab. The selected guard interval of the system is l/g_> FFT size SK, modulation 16-QAM and PP2 scattered pilots. The synchronization is assumed to be perfect during simulation. The carrier frequency is 862 MH2 and the sampling time t_s = 7/64 μς. As a channel we implement typical urban 6-path (TU-6) Rayleigh matlab channel model with a maximum Doppler shift of 200 Hz [11]. The maximum normalized Doppler shift will be about 0.21 based on f_i = fd_max/f_sc where f_sc is the subcarrier spacing equals to 977 Hz. In this scenario the symbol duration is 896 μ which can make the compensation very sensitive in time-variant channels to short coherence time. Here, it may be seen that the Doppler shift is nearly constant over one symbol duration.

Figure 4 shows the performance of the Doppler compensation in comparison to the l-tap frequency domain equalizer, BEM based equalizer (which is one of the state-of- art methods) and that without Doppler shift compensation case. The OMP outperforms all of them for all the SNRs. In other words, in comparison to 1- tap equalizer, it considers ICI explicitly which leads to a substantial improvement. In comparison to BEM model based scheme, it provides more accurate channel matrix which resulted with better compensation. It is important to note that OMP based equalizer is better especially for low SNR which makes it preferable.

Figure 5 shows the FEC threshold with red line after which BER becomes zero in case FEC is activated. The adaptive dictionary can reduce the power knowing that FEC will correct all bit errors starting from that FEC threshold. Hence, low precision dictionary can be used which requires less computation and BER which can be corrected by FEC. By using the adaptive dictionary, the computation can be reduced up to 45%. Here, FEC is concatenated LDPC and BCH codes in DVB-T2 with coding

Figure 6 introduces the adaptive stopping criterion which can reduce the computation after FEC threshold similar to the adaptive dictionary.

As the first stopping criterion (st.c.i) the algorithm is stoppend when T_p > 1 or R„ < is satisfied where for st.c.2 it will T„ > 1 or JL < ^■ and for It is important to mention that the bit errors do not occur due to the Doppler compensation algorithm, rather due to OMP algorithm. However, with FEC one does not need to lower the BER after given FEC threshold. MISO-OFDM SYSTEM

Figure 7 shows a MISO transmission system using multiple transmitters and one receiver, where the transmitted base data is the same in the transmitters but transmitted with slightly modified versions. In the receiver, y is the superimposition of them through different channels. The resulting diversity property improves the transmission.

The channel matrix for MISO can be written as follows

= (V ½ - ^hlN_t)

(7)

Here, M_t is the number of the transmitters. Output y for MISO can be written in a matrix form as

y(n) = H_MS(;n) · S(n) + w(n)

where S(n) is the transmitted signal vector.

Figure 8 illustrates a diversity scheme of the system based on the Alamouti methodology [6], [7]. The 7x1 and Tx2 are the transmitters which belong to MISO group 1 and group 2.

Thus, the transmitter in MISO group 1 transmits the signals without buffering, negation and complex conjugation. However, the signals transmitted by MISO group 2 transmitter undergo pairwise modification; Tx2 transmits —5_X ' while Txl transmits S₀, and Tx2 transmits S₀* while Txl transmits S_r The cells 5₀ and S₁ come from different Physical Layer Pipes (PLPs) and have no special relationship. In the receiver the signal components can be recovered accordingly where V_j and y₂ at time t and t + T can be written as ¾ = y(t) = h^So - fc^ + w_t y_z = y(t T Γ) = fe^S_j— .¾₁₂5, * JL

o where ii_tl and ¾₁₂. are the channel transfer functions between 7x1 and Rx and Tx2 and fi

After processing in the receiver, (8) can be written for MISO as

where:

One of the simple approaches to solving this problem is the simple matrix HMS inversion which is based on Zero-Forcing (ZF) and can be written as

Using (IO) data will be decoded back in the receiver as a result of the solution S.

After y has been reconstructed from the superimposed MISO signals, Doppler shifts and other channel effects can be compensated using the same methods as described in the single sender (SISO) case.

CHANNEL ESTIMATION

In the present embodiment, Alamouti-based MISO in frequency direction with two transmitters and a slightly modified version of the second transmitter's symbols is considered. The MISO processing and pilot modification for the signal from transmitters in MISO group l is unmodified. In contrast the group 2 is modified in the reverse order in frequency for each pair of the constellation. Since some pilots are inverted in MISO group 2 signal, the pilots for the same position can be calculated using the sum (sum) and the difference (diff) pilots. This is done by interpolation both in time and in frequency direction. After calculating sum and the diff pilots for the same positions one can find values for the channel l as (sum— diff)/ 2 and for the channel 2 as (sum + diffjjl. Finally, one can estimate both channels. According to figure 8, the channel coherence time, i.e. the time in which the Doppler shift is assumed to be constant, will need to be longer due to the interpolation in time - between symbols. Similarly, the density of the pilots will decrease twice in MISO in comparison to SISO. To achieve the same performance in the channel estimation, the number of the pilots should be increased twice.

If the channel matrix is unknown, it can be estimated using the methods described in connection with the single sender (SISO) case, using matching pursuit.

RESULTS FOR THE MISO-OFDM SYSTEM

The algorithms have been implemented in the DVB-T2 common simulations package (DVB-T2 CSP), [li]. The selected system parameters are MISO-OFDM with ¹/₈ guard interval, 8k mode, 256-QAM modulation and PP2 pilot symbols. Here, the synchronization and phase offset correction are assumed to be perfect; The carrier frequency and sampling interval are 862 MHz and t_s =■ 7/64μ5 respectively. As a channel Rayleigh MISO Channels have been adopted from [6], Tab. (1,11) ; The maximum Doppler shift is 200 Hz in the channel.

Figure 9 shows a comparison of different channel estimation algorithms, based on the bit error rate (BER). Here, we consider an overcomplete dictionary D with elements of 1 μβ time delay and 1 Hz Doppler shift difference. The LS-OMP algorithm outperforms the GRS based OMP and BMP algorithms and comes very close to the ideal channel information curve with a part per thousand precision. The GRS based OMP algorithm had a random deviation due to the coefficient estimation sensitivity.

Figure 10 compares the system without Doppler compensation (conventional) to the system with the proposed algorithm to demonstrate the Doppler compensation algorithm. The difference is very small between the ideal and the proposed one which indicates that the channel estimation is very accurate. Here, LS-OMP is used as a channel estimation algorithm.

Figure n compares MISO and SISO using LS-OMP and the proposed Doppler compensation. Here, MISO is better than SISO which is evident that MISO benefits the diversity property.

The Doppler shift under different acceleration rates a. If the symbol duration t_s is known one may calculate the Doppler shift difference between the beginning and the end of the symbol. First, the velocity difference Av between the beginning and the end of the symbol is calculated as Av = a - t_s then the Doppler difference is calculated as

Δ/¾ = Av · fo/c . Af_d will be 49m/s^■ (7/64 s^■ 862MH∑/c = 15 ΑμΗζ for the Formula.

One car, under heavy braking with the acceleration of 177 km/h². This shows that even in real high acceleration which corresponds to 177 km/s, the nonlinearity of the Doppler shift will be very small which is too far from making any negative effect on the compensation.

Claims

Method for compensating Doppler shifts (v) in a mobile communication system, comprising the steps of:

- receiving a mobile communication signal (y);

- transforming the signal in order to compensate a Doppler shift (v_p) of at least one propagation path (p) of the signal (y); and

- outputting the compensated signal.

Method according to claim 1 , wherein the step of transforming comprises applying a Fast Fourier Transform (FFT) to the signal.

Method according to claim 2, wherein the Doppler shift (v_p) is compensated based on an inverted channel matrix of the mobile communication system.

Method according to claim 3, wherein the channel matrix includes a representation of the Doppler shift (v_p) of the at least one propagation path (p) of the signal (y).

Method according to claim 4, further comprising the step of estimating the channel matrix from the received signal.

Method according to claim 5, wherein the channel matrix is estimated using a matching pursuit (MP) algorithm.

Method according to claim 6, wherein channel coefficients are determined using least squares (LS) minimization.

Method according to claim 7, wherein a stopping criterion for the matching pursuit algorithm takes a distance between channel coefficients for different numbers of paths into account.

Method according to claim 8, wherein a stopping criterion for the matching pursuit algorithm further takes a signal-to-noise-ratio (SNR) into account.

10. Method according to claim 4, wherein the channel matrix further includes a representation of at least one of the attenuation, phase shift and time delay of the propagation path.

11. Method according to claim 1, wherein the mobile communication signal is an OFDM signal.

OFDM receiver, adapted to carry out a method according to claim 1